A system to execute a process using a plurality of computation nodes in parallel is utilized. In the system, software called a scheduler executed by a node for management performs an allocation of processes to each computation node. For example, a technique has been suggested in which by a backfilling scheduler, jobs in a queue are placed on an unused space on a scheduler map in descending order of a priority and scheduled jobs are executed with the lapse of time (Japanese Laid-open Patent Publication No. 2012-215933). A technique has been suggested in which a small scale job is executed first without allowing the start of execution of a large scale job using many computing resources to be delayed by a backfilling scheduler and an operating ratio of a system is improved (Japanese Laid-open Patent Publication No. 2012-173753).
A technique has been suggested in which two types of schedules of a schedule in which temporal stopping of jobs being executed in a plurality of computing resources is allowed and the other schedule in which the temporal stopping is not allowed are prepared and a suitable schedule is determined by comparing a value indicating profits and a value indicating losses from both schedules (Japanese Laid-open Patent Publication No. 2013-41529 and International Publication Pamphlet No. WO 2012/020474)
A technique has been suggested in which better solutions are searched for, by a process called re-ordering, for solutions found by a priority algorithm in a method of solving a travelling salesman problem or a combinational problem such as job shop scheduling and the like (Japanese Laid-open Patent Publication No. 2005-56421).
As described above, a method is considered in which jobs are sequentially executed according to a priority at the earliest time at which each prioritized job may be executed. However, in this method, free time of computing resources is allocated simply to subsequent jobs according to the priority. For that reason, free time of computing resources is easily fragmentized to be segmented uselessly. However, there is a problem that simply allocating the jobs to the earliest time according to the priority is not necessarily efficient allocation of the computing resources to respective jobs from a viewpoint of scheduling result of a plurality of jobs. The time taken to complete all jobs may become longer.
One aspect of the present disclosure is to provide an information processing apparatus, a parallel computer system, a job schedule setting program, and a job schedule setting method that are capable of achieving an efficient job execution.